You are not registered yet. Please click here to register!
DFi UT x48-t3rs, X3360 - Page 18 - i4memory.com - different look at memory
Tags Register Blogs FAQ Members List Calendar Mark Forums Read

Notices

Reply
 
Thread Tools
Old 14-10-2011, 09:20 PM   #137
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
FSB, CPU and RAM goals were reached.

The Price:
1.325v on CPU
1.265v on VTT
1.43v on NB
1.93v on RAM.

For the elpidas still too high voltage, however it still reflects the Intels specs... Luckily no memory errors are measured...
Offler is offline   Reply With Quote
Old 17-10-2011, 11:18 PM   #138
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
PCI-E stress benchmark:

This tool sends data from CPU to GPU and then from GPU to CPU at various sizes up to 1gigabyte.

Best measured value was 6500mb/s (GPU>CPU) and 5500mb/s (CPU>GPU).

Why it is so few?

Possibilities.
1. Quad core architecture. Since benchmark is Singlethreaded it access only one core, L2 cache and possibly only half of LGA 775 bandwidth. Have to try it on E7200 to see.
2. these are real values and this is limit caused by current PCI-E 2.0 architecture.
3. CPU and GPU are too slow to generate such traffic.

Edit:
During the PCI-E test I tried running Everest. Read rates from memory lowered from 9500 to 8400 mb/s, but not less. This reflects only CPU load.by the benchmark test, but memory seems to work fine.

Last edited by Offler; 18-10-2011 at 11:58 PM.
Offler is offline   Reply With Quote
Old 02-12-2011, 08:06 PM   #139
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
Current configuration has changed a lot:

Memory 1600 CL 8-6-6-23 1,712v
NB 1,444v

System seems stable at first look. Each CPU based torture test will work fine, even 3d games are running fine.

Paradox is that system will most likely fail while playing UVD decoded streams = DVD playback and digital TV combined. This is causing quite high traffic on PCI-E, and memory respectively. Seems that traffic is higher than in any game. TV tuner application also takes 4gb of memory for its timeshift buffer.

This behaviour has only one reason - CPU is not able to utilize whole memory bandwidth. However multiple inputs and outputs from PCI-E bus are able to do so.
Offler is offline   Reply With Quote
Old 06-12-2011, 01:20 AM   #140
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
The new testing method for this combination of hardware is:

1) Start TV tuner application, select Satellite, open Full HD channel.
2) Start Windows Media Player. Start play DVD.
3) Start different TV tuner application. Start only DVB-T signal.
4) Start Linx with below normal.

The ProgDVB application start also create large timeshift buffer.

Under this load system has to fail in 30-60 minutes. However this also shows that previous memory and nb settings are not correct. Voltage back to 1,93v and I can start another round of testing.

Expectations are that system might be stable at 1,87v on memory and 1.33v on vNB...

some Intel documentation on PCI-E and CPU cooperation mechanism...
http://www.google.sk/url?sa=t&rct=j&...J29J8ZjimDtxBg

http://www.google.sk/url?sa=t&rct=j&...Inq-I0AkBHcV1Q

Last edited by Offler; 15-12-2011 at 10:31 PM.
Offler is offline   Reply With Quote
Old 15-12-2011, 11:04 PM   #141
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
Finally after reading all the intel and other documentation...

PCI-E device located directly on slot which is connected directly to Northbridge has possibility to read and write data without any action taken from the CPU.

PCI-E devices using PCI semantics (eg PCI cards in PCI-E slot using adapter) and all devices connected via DMI (southbridge), will allways need to snoop the transactions.

FSB snooping must be done to keep data in memory, CPU and PCI devices coherent. It is not necessary when device is reading/writing data which are not required by the CPU.

FSB snooping means that chipset "asks" cpu to check whether it has or not newer data in its cache(s) and update them as necessary. However even requests with NS bit (not snoop) may be snooped. Its not specified under which circumnstances. On Core 2 duo systems only one core is used, since the CPU has single cache. On Core2 quad, two of the CPUs are used, one for each physical cache.

This means that PCI-E graphic data (such as textures) may be transferred to the graphic memory (out and in) without taking any computing power from CPU. If the GPU is able to save data to RAM and read simultaneously at speed of 16gb/s total bandwidth of 25,6gb/s will be fully utilized if CPU is doing its transactions.

Snooped transactions will be limited by CPU host speed (in my case at 12,8gb/s) + time which takes the snooping itself.

Generally it means that DDR3 memory using both channels on 1600Mhz has its reason even on X48 and older chipsets, especially with both PCI-E 16x slots fully occupied by either GPUs or large SSD drives. It does not improve overall performance of system, but it improves "smoothness" of operations ocurring on same time from various sources and decreases "waiting" time on memory.

For example my OCZ Revo X2 drive uses PCI-X adapter and high speed PCI-X raid. even when connected to PCI-E slot it will always need FSB snooping.

And how this differs from DMA? Without dma, whole transferring is done with CPU, sacrificing a lot of its power only to data transport. With DMA only snooping is needed, while transfer is done standalone by the other device. Without snooping the data transfers are done fully on "hardware" level.

However even when snooping is applied the speed of memory at 1600mhz effective and FSB 1600Mhz Effective increases the speed of this action because data transfer from a single channel to cpu and to the PCI-E are limited only by the PCI-E and not by the memory or the CPU.

Generally 2000mb/s of this system's internal traffic must be snooped, rest - 16000 mb/s dont need to be snooped. However incorrectly chosen PCI-E slot might increase this rate of 11 percent up to 40 percent.

Last edited by Offler; 16-12-2011 at 12:00 AM.
Offler is offline   Reply With Quote
Old 09-01-2012, 10:24 PM   #142
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
So the cooling of NB has been set to a very special setting.

It keeps NB on temperatures between 51 and 63 degrees. Not cooler, but not hotter.

Voltages:
NB 1,42v
Memory 1,82v (still surviving elpida hypers).

Its strange but the system really hates temperatures below 42 degrees on NB. When the window is opened there is chance for an unplanned reboot. When the NB temperatures are higher there is also chance of reboot, but never when the temperatures are "average"...

Temperature obviously affect how the board works. Skew rates change the behaviour only on very minor level.

However reboots occur only when TV tuner application or 3d game (skyrim in this case) is running for a longer time of period in "unwanted" temperature.
Offler is offline   Reply With Quote
Old 13-01-2012, 09:52 PM   #143
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
Skew mode 0/0
Dram driving strenghts 1/3

Getting more heat to the CPU today. Since last change no bsod/crash.

Seems that "incorrect" temperatures and skew rates are affecting the stability more I expected. If possible i also try lower memory voltage...

Edit:
Issue with reboots finally identified. Happens when playing 3d game - Skyrim. Also when TV tuner is playing. Happens only after 10 or more hours of stable running, or in hot environment.

The issue occurs independenty on CPU stress. It seems also that memory stress does not cause the problem. Thermal dependance has been confirmed. NB and graphic card always involved...

So, NB will receive its fan again.

Edit2:
After another test when the situation repeated the system was "able" to fail every 30 minutes when tv tuner was running. I have set memory latencies manually to values recommended by XMP for 816Mhz and set tRD to 8. Following my own reccomendation to have it on same value as CAS. This fixed the problem. Today I will start tests with lower vNB and memory voltages.

Sadly performance of memory dropped...

Last edited by Offler; 19-01-2012 at 10:20 PM.
Offler is offline   Reply With Quote
Old 06-02-2012, 09:07 PM   #144
Offler
Senior Member
Offler's PC Specs
 
Join Date: Nov 18 2009
Location: Bratislava
Posts: 278
In later set of tests the issues with reboots were ocurring only when PCI-E was involved in the test. That means running of TV tuner, decoded via UVD.

All memory tests (like LinX) passed. So I added voltage on NB this time bit more as before (1,456v). All problems with reboots were gone. Even on 1.43v were still present on lower volts were gone.

For this problem I cannot blame quad core configuration, even not 8gb of ram. Seems like the NB core or PCI-E interface are the weakest spot. More likely it looks like problem of PCI-E, but here is any form of testing or overclocking very limited.
Offler is offline   Reply With Quote
Reply

Bookmarks

Tags
dfi, x3360, x48t3rs


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
 
Thread Tools


Similar Threads for: DFi UT x48-t3rs, X3360
Thread Thread Starter Forum Replies Last Post
Initial notes, findings and tips DFI UT X58-T3EH8 eva2000 DFI Intel Motherboard / CPU 0 04-08-2009 04:17 PM
[Bjorn3d.com] DFI Lanparty LT X48 T3RS News Reviews & News Online 0 09-12-2008 08:41 AM
DFI UT X48 T2R - C1 Error sunjovasie DFI Intel Motherboard / CPU 3 19-09-2008 01:17 AM
[Anandtech.com] DFI X48 LT T2R: Floats like a Butterfly… eva2000 Reviews & News Online 0 11-05-2008 02:14 AM
DFI STREET IS NO MORE ++++ NEW DFI CLUB FORUM IS UP!!!!!!!! dinos22 General Hardware 52 26-01-2007 10:17 PM


All times are GMT +11. The time now is 01:43 PM.

no new posts